Make connection update process timeout variable

[zacwbond]

This is a fix for #780, an issue I submitted earlier today.

I am connecting as a central to a remote peripheral using a connection interval of 120, latency of 90, and timeout of 3000. Due to the hard-coded 40 second timeout, it is impossible to perform a connection parameter update on this connection. Trying to do so results in a timeout followed by a disconnect 100% of the time.

My attempt to fix the problem is to compute the timeout based on the current connection parameters instead of just hard-coding a constant value.

[apache-mynewt-bot]

Style check summary

No suggestions at this time!

[rymanluk]

Thanks for rising the issue as this is valid one. We had some issues with this timer already before. However I would suggest some other solution for that issue.

Some background:
When controller does not support Connection Parameter procedure over LL, we are doing it over L2CAP. When we are doing it over L2CAP we delete this timer as soon as we receive over response L2CAP and not when actually parameters will changed. We did it because we had issues with timing as well and we decided to now wait for the controller to update parameters as it is actually his decision to update parameters or not. For some reason we keep timer in the case where Connection parameter procedure happens over LL but maybe it is time for change as it clearly can cause link drop when we don't want it.

So as said above, imho this timer shall not cause link termination when connection parameter update will not happen, because it is controller to decide if parameters shall be changed based on host suggestion.

I still think we need a timer to at least make sure that app does not send connection parameter updates too often, so I'm not saying we should remove it.

I would vote for different handling of this timer when it fires

a) remove ble_gap_update_notify(conn_handle, BLE_HS_ETIMEOUT); from ble_gap_update_timer(void) so we never notify app about timeout, but instead, when connection parameter finally happen, application will receive the event.

b) remove link termination on BLE_HS_TIMEOUT in ble_gap_update_notify

I would vote for a) @sjanc @andrzej-kaczmarek @ccollins476ad comments?

[zacwbond]

@rymanluk Your a) makes sense to me. If I get a timeout, I'd tend to assume that the event that timed out is never going to happen, so it would be odd to send a timeout event if the procedure is eventually going to succeed.

Another problem I've noticed is that I frequently see a call to my reset_cb with error code 19 when the gap connection parameter update procedure is initiated. I think it must be associated with the function ble_hs_hci_wait_for_ack(). That makes it sound like an HCI problem and not a Nimble problem, but it only ever happens with the parameter update procedure, and only intermittently.

[rymanluk]

@zacwbond could you update your PR so it implements a) ?

the other problem you mentioned is something with your controller I believe.

[zacwbond]

Closing because I mangled my forks somehow. Will re-open momentarily.